High-throughput sequencing
and genomes

Jelmer Poelstra

MCIC Wooster, OSU

2024-01-25

Sequencing technologies

DNA sequencing technologies: overview

  • Sanger sequencing (since 1977)
    Sequences a single, typically PCR-amplified, short-ish (≤900 bp) DNA fragment at a time

High-throughput sequencing (HTS)
Sequences 105-109 usually randomly selected DNA fragments (reads) at a time — two types:

  • Short-read sequencing
    • AKA Next-generation Sequencing (NGS)
    • Produces up to billions of 50-300 bp “reads”
    • Market dominated by Illumina
    • Since 2005 — technology stable

DNA sequencing technologies: overview

  • Sanger sequencing (since 1977)
    Sequences a single, typically PCR-amplified, short-ish (≤900 bp) DNA fragment at a time

High-throughput sequencing (HTS)
Sequences 105-109 usually randomly selected DNA fragments (reads) at a time — two types:

  • Short-read sequencing
    • AKA Next-generation Sequencing (NGS)
    • Produces up to billions of 50-300 bp “reads”
    • Market dominated by Illumina
    • Since 2005 — technology stable
  • Long-read sequencing
    • Reads much longer than in NGS but fewer, less accurate, and more costly per base
    • Two main companies: Oxford Nanopore Technologies (ONT) and Pacific Biosciences (PacBio)
    • Since 2011 — remains under rapid development

What about RNA?

This lecture mostly talks about DNA sequencing, but:

  • This includes the indirect sequencing of RNA after reverse transcription to cDNA, as in nearly all “RNA-seq”.

  • Direct RNA sequencing is possible with some of the sequencing technologies we discuss, as I’ll briefly mention later, but is hard and not (yet) widely used.


Similarly, the shorthand “sequencing”, like in “high-throughput sequencing” in the title of this presentation, generally refers to DNA sequencing.

Sequencing technology development timeline

Modified after Pereira et al. 2020 (www.ncbi.nlm.nih.gov/pmc/articles/PMC7019349/)

Sanger sequencing

Sequences a single, typically PCR-amplified, short-ish (≤900 bp) DNA fragment at a time.

Sequencing is performed by synthesizing a new DNA strand in part with fluorescently-labeled nucleotides — a different color for each base (A, C, G, T).1

The final result is a chromatogram that can be base-called:

https://dnacore.mgh.harvard.edu/new-cgi-bin/site/pages/sequencing_pages/seq_troubleshooting.jsp

A $100 million dollar genome: the entire human genome (3 Gbp) was sequenced with Sanger!

Sequencing cost through time

https://www.genome.gov/about-genomics/fact-sheets/Sequencing-Human-Genome-cost

Present-day Sanger applications

Common current applications of Sanger sequencing include:

  • Examining variation among individuals or populations in one or more candidate or marker genes (for population genetics, phylogenetics, functional inferences, etc.)

  • Taxonomic identification of a sample


Amplification of a target DNA fragment is usually done with PCR

  • This means that you need to (approximately) know in advance short flanking sequences to the sequence of interest — primers for your PCR.

  • Introns are good targets to sequence: variable sequences flanked by conserved sequences (exons) in which primers can be designed.

High-throughput sequencing applications

  • Whole-genome assembly

  • Variant analysis (for population genetics/genomics, molecular evolution, GWAS, etc.):

    • Whole-genome “resequencing”

    • Reduced-representation libraries (e.g. RADseq, GBS)

  • RNA-seq (transcriptome analysis)

  • Other functional sequencing methods like Methyl-seq, ChIP-seq, etc.

  • Microbial community characterization

    • Metabarcoding

    • Shotgun metagenomics

Two important variables in high-throughput sequencing:
read lengths & error rates

Read lengths

  • Short-read (Illumina) HTS: 50-300 bp reads

  • Long-read HTS: longer & more variable read lengths (PacBio: 10-50 kbp, ONT: 10-100 kbp)

When are longer reads useful?

  • Genome assembly

  • Haplotype and large structural variant calling

  • Transcript isoform identification

  • Taxonomic identification of single reads (microbial metabarcoding)

When does read length not matter (as much)?

  • Read-as-a-tag: the goal is just to know a read’s origin in a reference genome, like in counting applications such as RNA-seq

  • SNP variant analysis

Error rates

Currently, no sequencing technology is error-free, and several types of errors can occur:

  • Base call errors, e.g. a base that was called as an A may instead be a G.

  • Insertion or deletion (indel) errors

  • When the base calling software is not confident at all, it can also return Ns (= undetermined).

Quality scores in sequence data

When you get sequences from a high-throughput sequencer, base calls have typically already been made. Every base is also accompanied by a quality score (inversely related to the estimated error probability).

Overcoming sequencing errors

  • Sequencing at a high so-called “depth of coverage”1 generally allows to infer the correct sequence

  • But made more challenging by natural genetic variation among and within (heterozygosity due to diploid genomes) individuals

  • Typical depths of coverage: at least 50-100x for genome assembly; 10-30x for resequencing.

Illumina (short-read HTS / NGS)

  • 100-300 bp reads with 0.1-0.2% error rates

  • More reads, lower per-base cost, and lower error rates than long-read sequencing1.

  • Machines differ in throughput, cost per Gb, and read length:

Libraries and library prep

In a sequencing context, a “library” is a collection of nucleic acid fragments ready for sequencing.

In Illumina and other HTS libraries, the fragments number in the millions or billions and are often randomly generated from input such as genomic DNA — cf. Sanger sequencing:

The procedure that produces a library from a DNA (or RNA) extraction is called library prep.

Different library prep procedures are used depending on the type of sequencing (WGS, RAD-seq, RNA-seq, etc.) and HTS technology — and some include more specific fragment generation or selection. We’ll see the specific library prep steps for RNA-seq next week.

Libraries and library prep (cont.)

After library prep (here, for Illumina sequencing), each DNA fragment is flanked by several types of short sequences that together make up the “adapters”:



After talking about paired-end vs. single-end sequencing and the way Illumina sequencing works, we’ll take a closer look at the adapter components.

Paired-end vs. single-end sequencing

In Illumina sequencing, DNA fragments can be sequenced from both ends as shown below — this is called “paired-end” (PE) sequencing1:

When sequencing is instead single-end (SE), it simply means that there is no reverse read:

Paired-end sequencing

  • Paired-end sequencing is a way to effectively increase the read length. (In the resulting sequence files, the two reads in each pair are separate, but can be matched thanks to shared read IDs.)

  • Earlier, we saw that the maximum read length of Illumina is 300 bp but in paired-end sequencing, this becomes “2 x 300 bp”, etc.

  • The total size of the biological DNA fragment (without adapters) is often called the insert size:

Insert size variation

The insert size varies — because the library prep protocol can aim for various sizes, and because of variation due to limited precision in size selection. It can be:

  • Shorter than the combined read length, leading to overlapping reads:

  • Shorter than the single read length, leading to “adapter read-through”
    (i.e., the ends of the resulting reads will consist of adapter sequence):

How Illumina sequencing works

First, library fragments bind to a surface thanks to the adapters, and the DNA templates (the biological sequences) are then PCR-amplified to form “clusters” of identical fragments:

In the diagram above, for illustrative purposes:

  • Only a few nucleotides are shown (1 block = 1 nucleotide) — in reality, fragments are much longer

  • Only two templates => clusters are shown — in reality, there are millions

How Illumina sequencing works (cont.)

Then, sequencing is performed by synthesizing a new strand using fluorescently-labeled bases and taking a picture each time a new nucleotide is incorporated:

How Illumina sequencing works (cont.)

How errors come about in Illumina

  • The different templates within a cluster get out of sync because occasionally:

    • They miss a base incorporation

    • They incorporate two bases at once

  • Base incorporation may also terminate before the end of the template is reached

This error profile is why, for Illumina:

  • There are hard limits on read lengths

  • Base quality scores decrease along the read

How Illumina sequencing works: Zooming out

How Illumina sequencing works: Zooming out

A closer look at the adapter components

Now that you have a better idea of how Illumina sequencing works, let’s take a closer look at the different components of the adapters flanking the DNA:


Modified after http://nextgen.mgh.harvard.edu/IlluminaChemistry.html

Long-read HTS

The technologies underlying the two main long-read HTS technologies are very different, but have some commonalities beyond long reads — they:

  • Perform “single-molecule” sequencing (no PCR amplification of library fragments)
  • Require higher quality & quantity DNA
  • Can detect some base modifications, like methyl groups
  • Can sequence RNA directly

Error rates are changing

As a shorthand that was universally true until recently, I mentioned earlier that long-read HTS has higher error rate than short-read (Illumina) HTS.

However, error rates in one type of PacBio sequencing where individual fragments are sequenced multiple times (“HiFi”) are now lower than in Illumina.

Nanopore sequencing

A single strand of DNA passes through a nanopore —
what is measured is the electrical current, which depends on which combination of bases passes through:

https://www.genome.gov/genetics-glossary/Nanopore-DNA-Sequencing

ONT (Nanopore) sequencers

Under development!

ONT constantly releases new flow cells with updated technology, which have led to large decreases in error rates over the past decade and even over the past two or so years.

At the same time, there is also a lot of development in the base-calling software. Unlike with Illumina or PacBio, it is common & useful to receive raw, not-basecalled data files: re-basecalling the same data a few years later with new software (versions) can make a substantial difference.

ONT vs. PacBIO

Advantages of ONT:

  • Low capital cost, portability (in-the-field sequencing!)

  • Read length not inherently limited, some extremely long reads

  • Lower cost per base


Disadvantages of ONT:

  • Higher error rates

  • Some systematic errors (e.g. homopolymers)

Genomes

Genomes

As methods facilitating genomics and transcriptomics research, genomes loom large in HTS. Specifically, most HTS applications either require a “reference genome” or involve its production.

What exactly does “reference genome” refer to? Three components to this phrase:

  • Assembly
    It includes a representation of most of the genome DNA sequence: the genome assembly
  • Annotation
    It (preferably) includes an “annotation” that provides the locations of genes and other genomic features, as well as functional information on these features

  • Taxonomic identity
    This is typically considered at the species level, in which case it should involve the focal species. But:

    • If necessary, it is often possible to work with reference genomes of closely related species

    • Conversely, different reference genomes may exist for different subspecies/populations within a species

Genome size variation

https://en.wikipedia.org/wiki/Genome_size

Genome structure

https://en.wikipedia.org/wiki/Karyotype





Key features:

  • Number of distinct chromosomes

  • Ploidy

Genome assembly & annotation

Creating a reference genome has two main steps: assembly and annotation.


A few notes on genome assembly:

  • Most assemblies are not “chromosome-level”, i.e. they don’t have one contiguous sequence for each chromosome. Instead, they consist of contigs and scaffolds, which can number in the thousands.

  • Even chromosome-level assemblies are not 100% complete

  • With increasing usage & quality of long-read HTS, we are generating better assemblies

  • To create chromosome-level assemblies, other technologies typically also needed (Hi-C, optical mapping)

Genome assembly & annotation

A few notes on genome annotation:

  • The first step is structural annotation: the identification of genes and other genomic features within the genome sequence

  • The second step is functional annotation: giving names and assigning functions to (mostly) genes

  • Whereas genome assembly typically does not borrow information from other organisms, genome annotation very heavily relies on that, based on the concept of sequence homology.


How is this data stored?

Both genome assemblies and annotations are typically saved in a single text file each — more on that soon.

Appendix: Sequence data files

Overview

All common genetic/genomic data files are plain-text, meaning that they can be opened by any text editor. However, they are often compressed to save space. The main types are:

  • FASTA
    Simple sequence files, where each entry contains a header and a DNA/AA sequence.
    Versatile, anything from a genome assemblies, proteomes, and single sequence fragments to alignments can be in this format.

  • FASTQ
    The standard format for HTS reads — contains a quality score for each nucleotide.

  • SAM/BAM
    An alignment format for HTS reads

  • GTF/GFF
    Tables (tab-delimited) with information such as genomic coordinates on “genomic features” such as genes and exons. The files contain reference genome annotations.

FASTA files

FASTA files contain one or more (sometimes called multi-FASTA) DNA or amino acid sequences, with no limits on the number of sequences or the sequence lengths.

As mentioned, they are versatile, and are the standard format for:

  • Genome assembly sequences

  • Transcriptomes and proteomes (all of an organism’s transcripts & amino acid sequences, resp.)

  • Sequence downloads from NCBI such as a single gene/protein or other GenBank entry

  • Sequence alignments (but not from HTS reads)

FASTA files (cont.)

The following example FASTA file contains two entries:

>unique_sequence_ID Optional description
ATTCATTAAAGCAGTTTATTGGCTTAATGTACATCAGTGAAATCATAAATGCTAAAAA
>unique_sequence_ID2
ATTCATTAAAGCAGTTTATTGGCTTAATGTACATCAGTGAAATCATAAATGCTAAATG

Each entry contains a header and the sequence itself, and:

  • Header lines start with a > and are otherwise “free form” but usually provide an identifier (and sometimes metadata) for the sequence
  • A single sequence is often not on a single line, but spread across multiple lines with a fixed width

FASTA file name extensions are variable:

  • Generic extensions are .fasta and .fa

  • Also used are extensions that explicitly indicate whether sequences are nucleotide (.fna) or amino acids (.faa)

FASTQ

FASTQ is the standard format for HTS reads.
Each read forms one FASTQ entry and is represented by four lines, which contain, respectively:

  1. A header that starts with @ and e.g. uniquely identifies the read
  2. The sequence itself
  3. A + (plus sign)
  4. One-character quality scores for each base (hence FASTQ as in “Q” for “quality”)

FASTQ quality scores

The quality scores we saw in the read on the previous slide represent an estimate of the error probability of the base call.

Specifically, they correspond to a numeric “Phred” quality score (Q), which is a function of the estimated probability that a base call is erroneous (P):

Q = -10 * log10(P)

For some specific probabilities and their rough qualitative interpretation for Illumina data:

Phred quality score Error probability Rough interpretation
10 1 in 10 terrible
20 1 in 100 bad
30 1 in 1,000 good
40 1 in 10,000 excellent

FASTQ quality scores (cont.)

This numeric quality score is represented in FASTQ files not by the number itself, but by a corresponding “ASCII character”.

This allows for a single-character representation of each possible score — as a consequence, each quality score character can conveniently correspond to (& line up with) a base character in the read.

Phred quality score Error probability ASCII character
10 1 in 10 +
20 1 in 100 5
30 1 in 1,000 ?
40 1 in 10,000 I

A rule of thumb

In practice, you almost never have to manually check the quality scores of bases in FASTQ files, but if you do, a rule of thumb is that letter characters are good (Phred of 32 and up).

FASTQ (cont.)

FASTQ files have no size limit, so you may receive a single file per sample, although:

  • With paired-end (PE) sequencing, forward and reverse reads are split into two files:
    forward reads contain R1 and reverse reads contain R2 in the file name.

  • If sequencing was done on multiple lanes, you get one (SE) or two (PE) files per lane per sample.1

FASTQ files have the extension .fastq or .fq (but are commonly compressed, leading to fastq.gz etc.). All in all, having paired-end FASTQ files for 2 samples could look like this:

# A listing of (unusually simple) file names:
sample1_R1.fastq.gz
sample1_R2.fastq.gz
sample2_R1.fastq.gz
sample2_R1.fastq.gz

GTF/GFF

The GTF and GFF formats are tab-delimited tabular files that contain genome annotations, with:

  • One row for each annotated “genomic feature” (gene, exon, etc.)

  • One column for each piece of information about a feature, like its genomic coordinates

See the sample below, with an added header line (not normally present) with column names:

seqname     source  feature start   end     score  strand  frame    attributes
NC_000001   RefSeq  gene    11874   14409   .       +       .       gene_id "DDX11L1"; transcript_id ""; db_xref "GeneID:100287102"; db_xref "HGNC:HGNC:37102"; description "DEAD/H-box helicase 11 like 1 (pseudogene)"; gbkey "Gene"; gene "DDX11L1"; gene_biotype "transcribed_pseudogene"; pseudo "true"; 
NC_000001   RefSeq  exon    11874   12227   .       +       .       gene_id "DDX11L1"; transcript_id "NR_046018.2"; db_xref "GeneID:100287102"; gene "DDX11L1"; product "DEAD/H-box helicase 11 like 1 (pseudogene)"; pseudo "true";

Some details on the more important/interesting columns:

  • seqname — Name of the chromosome, scaffold, or contig
  • feature — Name of the feature type, e.g. “gene”, “exon”, “intron”, “CDS”
  • start & end— Start & end position of the feature
  • strand — Whether the feature is on the + (forward) or - (reverse) strand
  • attribute — A semicolon-separated list of tag-value pairs with additional information

SAM/BAM

Using specialized bioinformatics tools, you can align HTS reads (in FASTQ files) to a reference genome assembly (in a FASTA file).

The resulting alignments are stored in the SAM (uncompressed) / BAM (compressed) format.

SAM/BAM are tabular files with one line per alignment, each of which includes:

  • The position in the genome that the read aligned to

  • A mapping score based on the length of the alignment and the number of mismatches

  • The sequence of aligned the read itself

File conversions

  • FASTQ files can be converted to FASTA files (losing quality information) but not vice versa

  • SAM/BAM files can be converted to FASTQ files (losing alignment information) but not vice versa

  • Proteome FASTA files can be produced from the combination of a FASTA genome assembly and a GFF/GTF genome annotation